Cross-subject Error-related EEG Signal Detection Based on Riemannian Manifold and Filter Bank

Brain-computer interface (BCI) attracted more and more attention in the past decade, and research about EEG signal has increased gradually. However, due to the large variability and volatility of different subjects EEG signals, cross-subject models often perform poorly. This paper proposes a method to identify error-related potential (ERRP) P300 signals which come from public dataset. First, using the XDAWN algorithm to estimate the average evoked potentials and add them to the original signals. By this way, making the Riemannian mean of difference classes farther. Second, selecting electrode channels based on the Riemannian distance of Riemannian mean, reducing convergence time effectively. Third, extracting Riemannian features based on Riemannian kernel. Finally, combining multi-band features and projecting features of different domains into the common feature space, and then detecting by a suitable way. The method proposed in this experiment will be compared with the other methods. And we will analyze the results to prove the effectiveness in the situation that uses a single subject as the target domain.